ENAS

E899017

AutoML technique neural architecture search method

ENAS (Efficient Neural Architecture Search) is a method that dramatically reduces the computational cost of neural architecture search by sharing parameters among many candidate architectures within a single super-network.

Try in SPARQL Jump to: Statements Referenced by

Statements (47)

Predicate	Object
instanceOf	AutoML technique ⓘ neural architecture search method ⓘ
abbreviationFor	Efficient Neural Architecture Search NERFINISHED ⓘ
aimsTo	reduce computational cost of neural architecture search ⓘ
appliedTo	image classification ⓘ language modeling ⓘ
assumes	weight sharing does not overly bias architecture evaluation ⓘ
benchmarkDataset	CIFAR-10 NERFINISHED ⓘ Penn Treebank NERFINISHED ⓘ
category	meta-learning method ⓘ model search algorithm ⓘ
citationType	highly cited NAS paper ⓘ
comparedTo	NASNet NERFINISHED ⓘ Neural Architecture Search with Reinforcement Learning NERFINISHED ⓘ
controllerOutput	architecture decisions ⓘ
domain	deep learning ⓘ
evaluationMetric	validation performance of sampled architectures ⓘ
field	artificial intelligence ⓘ machine learning ⓘ
fullName	Efficient Neural Architecture Search NERFINISHED ⓘ
hasTitle	Efficient Neural Architecture Search via Parameter Sharing NERFINISHED ⓘ
improvesOver	standard neural architecture search in efficiency ⓘ
influenced	later efficient NAS methods ⓘ
optimizesFor	computational efficiency ⓘ validation accuracy ⓘ
organizationAffiliation	Google Brain NERFINISHED ⓘ
proposedBy	Barret Zoph NERFINISHED ⓘ Hieu Pham NERFINISHED ⓘ Jeff Dean NERFINISHED ⓘ Melody Y. Guan NERFINISHED ⓘ Quoc V. Le NERFINISHED ⓘ
publicationYear	2018 ⓘ
publishedIn	arXiv NERFINISHED ⓘ
reduces	GPU hours required for architecture search ⓘ search time by orders of magnitude compared to earlier NAS methods ⓘ
samples	subgraphs from a super-network ⓘ
searchesOver	neural network architectures ⓘ
searchGranularity	cell-level architecture search ⓘ
searchSpaceType	cell-based search space ⓘ
searchStrategy	RL-based controller over shared-weights supernet ⓘ
sharesParametersAmong	candidate architectures ⓘ
superNetworkType	directed acyclic graph ⓘ
trains	a single super-network ⓘ
uses	controller RNN ⓘ parameter sharing ⓘ reinforcement learning ⓘ
usesOptimization	policy gradient ⓘ

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

Neural Architecture Search → notableMethod → ENAS ⓘ